HeyGen introduced Avatar V in a dated 9 September 2026 guide as a digital-twin workflow based on a short video of a real person.
A motion reference is personal data
It asks for about 15 seconds of expressive motion capture and offers the ability to apply that motion identity to approved photo looks. This is more than a facial image animation: it is the reuse of someone’s characteristic gestures and delivery. A production team should therefore handle the reference footage as identity-bearing media with documented permission and access restrictions.
What the model guide explicitly differentiates
The launch document emphasizes real-human digital twins for Avatar V and virtual, nonhuman or stylized characters for Avatar IV. It specifies that Avatar V is tied to a video-based Look, and its published Studio rate is 48 credits per output minute for that Look. The provider also describes deployment through Studio, a homepage shortcut and Video Agent. These details should not be generalized into an API price or a universal enterprise entitlement.
Capture a reproducible Look
Get written permission that names the allowed languages, channels, project and retention period. Record the 15-second reference in even light with meaningful facial expression, natural gestures and no unlicensed soundtrack. Use a clear approved base portrait; avoid changing age or other identity-defining characteristics without separate permission. A script should identify pronunciation, pauses and any claims that need human review. Store the permission reference with the generated media, but do not publish legal records in an article export.
Build for accuracy rather than novelty
For a two-minute customer-onboarding video, split the script into clear steps and generate a first draft in the editor. Review lip timing and posture around difficult names, company acronyms and gestures. Compare the synthesized presenter with the person’s stated preferences; plausible movement is not the same as authorized movement. Require a designated approver before external distribution. Where a brand uses a fictional narrator instead, consider an Avatar IV workflow and label the figure appropriately.
Commercial and technical caveats
HeyGen states an up-to-three-minute Avatar V use within Video Agent, but that sentence does not establish identical caps in every integration. Rates and entitlements differ between web, API and legacy plans. Vendor descriptions of realism are promotional, not controlled measurements. For 4K deliverables, check whether an API produces native detail or merely upscales the avatar component. A consented digital twin should have a revocation process and a documented audit trail.
Relevant alternatives and trade-offs
Avatar IV may be more appropriate for nonhuman presenters. A traditional camera shoot avoids synthesized identity reuse entirely; for lip-synced still characters, Hedra Character 3 has a separate audio-driven specification.